No. of Recommendations: 8
The point is that - unlike x-rays or MRIs in radiology - tests for automated diagnosis of serious cardiac conditions can be done without AI, without excessive computational cost, without expensive and scarce equipment, with low patient risk, and without a trained technician. And this is *not being done*. So why not? And why would AI change this?
As you correctly point out, many things can be done by regular programming OR AI. In some theoretical sense, anything that can be done by AI can be done by regular programming, except regular programming might require VASTLY (read exponentially) more programming resources than will AI. For all intents and purposes, AI automates an extremely large and complex part of programs that deal with gigantically complex systems with many moving parts all interacting. AI automates the production of programs which is essentially the same as saying AI automates the production of software.
So in some sense, the question is why would you use one of AI or regular programming? I think the answer is paradoxically two fold:
1) You will use regular programming for incredibly simple stuff, a small number of interacting variables with reasonably well understood interactions
2) You will ultimately use AI to write the regular programs for when you need them!
In some sense, your question is similar to a question that might have been asked 100 years ago. Horses can bring the milkman's milk from house to house perfectly well with no new expensive motorized trucks. So why not? Why would we switch to trucks? And certainly for some while, horses did continue to be used for many hauling and transportation tasks, but ultimately motorized vehicles became the preferred technology for 99.99999% of transportation needs.
And it will be like that with AI. A detailed study could attempt to predict how long it will take AI to spread through the economy, taking over the jobs it will take over. But without doing that study, I feel pretty comfortable saying it will be years to decades in the US, and possibly 100 years for the planet as a whole. The extremely large computing centers needed to train AI will limit early AI deployment only to the most valuable tasks. I don't know what those will be, but safe to say "AI assistants" and "AI Driving" will be early applications, based on the fact that those two are already well along the path of development. General purpose, probably humanoid robots will likely be relatively early, but likely at least a year away before we see any seriously commercial deployment of these. But then realizing that literally billions of these robots would have to be built before they are used for the lowest valued jobs done by humans, and how many robot factories have to be built to accomplish that, and how long does that buildout take? Some idea of the answer to that: Tesla has been mass producing cars since 2013, and has currently reached only 3 million cars on the road, after 11 years. Even at 25% growth per year, it would take another 25 years to get to 1 billion cars. So many billions of robots serving man could easily take 30 years or more.
But the days of writing large complex software without using AI to do most of that work are essentially already over. As soon as someone figures out some improvements they want to make to the processing of X-rays, this will NOT be done by improving the already existing codebase by hand, but rather the existing codebase will help produce a part of the traning set for an AI implementation. And the AI implementation will handle corner-cases better than the hand=programmed version.
R:)
